An Interconnection Network Based on Asynchronous Optical Packet Switching for High Performance Computing System

摘要

A novel interconnection network based on asynchronous optical-packet-switching and optical multicast for high performance computing system applications is presented. Distributed management multi-stage structure is used to alleviate the storage stresses and arbitration complexities for every stage and is easy to realize all-optical scalability. Asynchronous switching mode is adopted for processing momentarily the burst requests from any CPUs interconnected to the node, and optical multicast technology is used to improve the the network switching efficiencies. A packet retransmission based collision resolution algorithm is proposed so as to be suitable for transmitting the burst services. The simulation models for the blocking rates, packets waiting delays, stability features and so on in the network are set up, which have been approved to be efficient for the two-stage system interconnecting 6,400 CPUs in the light of using dense-wavelength-division-multiplexing (DWDM) technology with 80 wavelength channels and capacity of 40 Gbit/s per channel.